Lesson 4: Machine Learning

Question 1

Accuracy (classification)

Answer 1

A measure which is defined as the number of correct predictions divided by the total number of predictions.

Question 2

Area Under the Curve

Answer 2

the percentage of area underneath the ROC curve. This is a measure of how accurate the two-class model is, with numbers closer to 1 being better.

Question 3

Azure Machine Learning Studio

Answer 3

The integrated development environment (IDE) for Azure Machine Learning.

Question 4

Compute (Azure ML)

Answer 4

Virtual machine resources which are dedicated to performing tasks in Azure Machine Learning. Compute may include individual virtual machines (VMs), typically configured as data science VMs, or it may include a cluster of VMs intended for training and inference pipeline executions.

Question 5

Confusion matrix:

Answer 5

A table representing predicted versus actual values for a classification problem. A classic two-class confusion matrix has four boxes. Using “Yes” and “No” as the two classes, these four boxes are:

Question 6

How many types of values are there in a confusion matrix?

Answer 6

four

Question 7

Name the four confusion matrix types

Answer 7

True Positive: we predicted Yes correctly
False Positive: we predicted Yes but it was really No
False Negative: we predicted No but it was really Yes
True Negative: we predicted No correctly

Question 8

True Positive

Answer 8

from the confusion matrix: we predicted Yes correctly

Question 9

False Positive

Answer 9

from the confusion matrix: we predicted Yes but it was really No

Question 10

False Negative

Answer 10

from the confusion matrix: we predicted No but it was really Yes

Question 11

True Negative

Answer 11

from the confusion matrix: we predicted No correctly

Question 12

Data Labeling

Answer 12

This functionality allows you to label images as part of an image classification project.

Question 13

Experiment (Azure ML)

Answer 13

A collection of trials used to validate a user’s hypothesis. An experiment may contain multiple runs of pipelines.

Question 14

Feature

Answer 14

Inputs which help us understand what affects the label.

Question 15

Feature engineering

Answer 15

Creating new features from existing data. This might include calculating new features, translating a street address into latitude and longitude, or parsing passages of text for meaning.

Question 16

Feature selection

Answer 16

Removing a column from consideration when training a model.

Question 17

Label

Answer 17

The thing we want to predict.

Question 18

Linked Services

Answer 18

This functionality allows you to integrate Azure Machine Learning with other Azure services. At present, the only linked service offering is to connect to Azure Synapse Analytics, which is a modern data warehousing offering on Azure.

Question 19

Mean Absolute Error (MAE)

Answer 19

An evaluation measure for any regression model. It is the average difference between predicted and actual values. This works well when dealing with small ranges of numbers.

Question 20

Mean Absolute Percent Error (MAPE)

Answer 20

An evaluation measure for any regression model. It is the percentage difference between the predicted and actual values. If the actual value is 0, MAPE will fail with a divide by 0 error, so it is not a good measure if the actual value can be 0. MAPE works best when you have large ranges of numbers.

Question 21

Microservice

Answer 21

A lightweight, independent service. Typically, microservices have one job and communicate with each other using well-defined operations.

Question 22

Node (input, output)

Answer 22

An input or output connection point on a component. Each component will have 0 to 3 input nodes and 0 to 3 output nodes. Each input or output node has a specific type, such as DataFrameDirectory, TransformationDirectory, or UntrainedModelDirectory. An input of DataFrameDirectory can only attach to an output of the same type.

Question 23

Overfitting

Answer 23

A situation which happens when a trained model latches onto the particular relationships within a training data set, but those particulars are not always indicative of the broader world.

Question 24

Pipeline (Azure ML)

Answer 24

A collection of components connected together in a defined order. The metaphor represents how data moves from a source (an initial dataset) and flows through components until it reaches a destination. There are two types of pipeline: training pipelines and inference pipelines.

Question 25

Pipeline Asset

Answer 25

A component available within Azure Machine Learning. This includes datasets you have imported, sample datasets which come with the service, and different components to transform, train, evaluate, and deploy models.

Question 26

Precision

Answer 26

A measure which calculates how frequently our predicted value is correct. It is defined as True Positives / (True Positives + False Positives).

Question 27

R^2 (R-squared)

Answer 27

An evaluation measure for linear regression models which ranges from 0-1, where 1 is the highest possible score.

Question 28

Recall

Answer 28

A measure which calculates how frequently we correctly predict a value. It is defined as True Positives / (True Positives + False Negatives).

Question 29

Receiver Operating Characteristic (ROC) curve

Answer 29

A plot which represents true positive versus false positive rates for a two-class model.

Question 30

Reinforcement learning

Answer 30

A machine learning technique in which we train an agent to observe its environment and use those environmental clues to make a decision.

Question 31

Root Mean Square Error (RMSE)

Answer 31

An evaluation measure for any regression model. RMSE works best when you are concerned with large differences between the predicted and actual values.

Question 32

Root Mean Square Log Error (RMSLE)

Answer 32

An evaluation measure for any regression model. RMSLE works best when you are concerned with large percentage differences between the predicted and actual values.

Question 33

Run (Azure ML)

Answer 33

An attempt to train a model in Azure Machine Learning. This can be done through a pipeline in the Azure ML designer or through Automated ML.

Question 34

Semi-supervised learning

Answer 34

A machine learning technique in which we have a small percentage of data with labels and a large percentage of unlabeled data.

Question 35

Sink node

Answer 35

A node with no outputs. An example of a sink node is Web Service Output.

Question 36

Source node

Answer 36

A node with no inputs. An example of a source node is any dataset you bring onto the canvas.

Question 37

Supervised learning:

Answer 37

A machine learning technique in which we have a known good answer for our label and attempt to learn from this label for inference purposes. The most common examples of this include classification and regression.

Question 38

Unsupervised learning

Answer 38

A machine learning technique in which we do not have labels for our data. We use unsupervised learning techniques to try to discover what those labels should be. Clustering is the most common example of this.